Combinatorial Multi-Armed Bandits with Fairness Constraints: An Online Convex Optimization Perspective
Authors: Xiaosong Chen, Hanqin Zhuang, Yang Liu, Huanle Xu, Wing Cheong Lau
JAIR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we assess the performance of our algorithm through extensive simulations and real dataset applications, demonstrating its significant advantages over baseline schemes. In this section, we assess the performance of the CMF and CMFK algorithms separately. We detail the simulation results for the CMF algorithm in Section 7.1 and for the CMFK algorithm in Section 7.2. Furthermore, we explore real-world applications in Section 7.3. |
| Researcher Affiliation | Academia | Xiaosong Chen EMAIL Department of Computer and Information Science University of Macau Macau, China; Hanqin Zhuang EMAIL Department of Computer and Information Science University of Macau Macau, China; Yang Liu EMAIL Department of Computer Science Shanghai University Shanghai, China; Huanle Xu (Corresponding Author) EMAIL Department of Computer and Information Science University of Macau Macau, China; Wing Cheong Lau EMAIL Department of Information Engineering The Chinese University of Hong Kong Hong Kong, China |
| Pseudocode | Yes | Algorithm 1: Combinational Multi-arm Selection Algorithm with Fairness Guarantees; Algorithm 2: Combinational Multi-arm Selection Algorithm with Fairness Guarantees and Knapsack constraints |
| Open Source Code | No | The paper does not provide any explicit statement or link indicating that the source code for the described methodology is publicly available. |
| Open Datasets | Yes | To conduct our experiments, we utilize the Movie Lens 20M Dataset (Harper & Konstan, 2015), which comprises 20 million ratings for 27,000 movies provided by 138,000 users. |
| Dataset Splits | No | The paper mentions using the Movie Lens 20M Dataset and selecting 100 movies and 1500 users as rounds, but it does not specify how this data was split into distinct training, validation, or test sets in a traditional sense. The nature of a bandit problem implies sequential processing rather than static splits. |
| Hardware Specification | Yes | Specifically, CMF only requires 32ms to select arms in each round on an Intel i9 core processor. |
| Software Dependencies | No | The paper does not provide specific version numbers for any software libraries, frameworks, or programming languages used in the experiments. |
| Experiment Setup | Yes | We consider the following scenario for the simulation: N = 100 and m = 30. The values of ξ are generated uniformly at random between [0.01, 1] and PN i=1 ξi = 15. The expected reward for all arms are uniformly chosen between [0,1]. For each arm, the actual rewards in all rounds are generated following the Pareto distribution with the order of two. In this experiment, we choose the reward function to be linear. ... We consider the scenario for simulation: N = 100 and B = 2000. For each arm, the actual rewards in all rounds are generated following the Pareto distribution whose expectations uniformly drawn between [0,1]. And the actual resource consumption in all rounds is generated following the uniform distribution whose expectations are drawn between [0,0.2]. ... In particular, we choose linear, power, and logarithmic functions as representatives. The specific forms of these functions are f(y) = y, f(y) = log(y + 1), and f(y) = y + 1. ... Additionally, we set the fairness constraint to be 0.03 across all experiments. |